AI Exposures in Insurance: Emerging Risks with Verisk CEO Lee Shavel

Understanding the risk of AI starts with knowing where it shows up. 🔍   From underwriting to autonomous vehicles, artificial intelligence is reshaping the insurance landscape in real time.   Our CEO, Lee Shavel, breaks down where the exposures are emerging and what insurers need to do to stay ahead.   Read the full article here: https://ow.ly/4VlP30sVMi0   #EmergingRisks #AI #Insurance

Understanding the risk of AI starts with knowing where it shows up. 🔍   From underwriting to autonomous vehicles, artificial intelligence is reshaping the insurance landscape in real time.   Our CEO, Lee Shavel, breaks down where the exposures are emerging and what insurers need to do to stay ahead.   Read the full article here: https://ow.ly/pOfR30sVKg6   #EmergingRisks #AI #Insurance

Understanding the risk of AI starts with knowing where it shows up. 🔍   From underwriting to autonomous vehicles, artificial intelligence is reshaping the insurance landscape in real time.   Our CEO, Lee Shavel, breaks down where the exposures are emerging and what insurers need to do to stay ahead.   Read the full article here: https://ow.ly/mHJt30sVKzY   #EmergingRisks #AI #Insurance

In ninety-eight days, EU AI Act Article 9 starts being enforced for every high-risk AI system on the European market. Article 9 requires a continuous risk management system. Identify the foreseeable harms. Evaluate them in real conditions. Document the result. Hold up under examination. The article describes what must be documented. It does NOT say what the document looks like. No format. No template. No agreed receipt. We built one. Then we tested it. We ran our governor against real autonomous vehicle data, about two million moments of actual driving. The car's own controller said the drive was perfect. One hundred percent normal, the whole way through. Our governor disagreed about one in five moments. It also caught five thousand instances where the car switched from "everything is fine" to "stop now" without flagging it. Same vehicle. Same drive. Two different stories about whether it was safe. We have the data. If your firm has a high-risk AI system on the European market, the question is not whether you will be examined. It is what your evidence file will say about the system that was actually running, versus the one that wrote its own report. JonathanLuethke@WayfinderSystemsGroup.com #AIGovernance #EUAIAct #ModelRiskManagement #AIRiskManagement

The story is simple. When an AI makes its own decisions, you cannot ask the AI whether it was safe. I have watched the same pattern surface in autonomous vehicles, insurance, lending, claims, healthcare triage. Every adaptive system grades itself. To what standard? In ninety-eight days the EU stops accepting that as an answer. The NAIC and the OCC already do not. Auditors are signing off on evidence files that describe systems which no longer exist. The boards above them are too. So we built a runtime governor and we built the receipt format that comes out of it. One record, three readers. Regulator, auditor, operator. We tested it against real autonomous vehicle data. The car's controller said the drive was perfect. Our governor caught one in five moments where it was not.

Autonomous vehicles are not failing because they can’t see. They’re failing because no one owns the moment they act. Every system today is built to answer: “What is the best possible decision?” But that isn’t the real question. The real question is: At the exact moment a vehicle acts, what proves it was safe to act? Not: – what it saw – what it calculated – what it intended But: what evidence existed, in that moment, that made the action admissible Right now, that doesn’t exist. We have: – perception – planning – control But no mechanism that determines whether execution is allowed. Which means: Vehicles don’t fail because they make the wrong decision. They fail because they are allowed to act without proving they should. Until that changes, autonomy will always rely on reconstruction after the fact. Not proof at the moment it matters. So the question is simple: What determines whether an autonomous system is allowed to act at all? #AutonomousVehicles #AI #SafetyEngineering #SystemsEngineering #RiskManagement #FunctionalSafety #Governance

What Our Fear of AI Reveals About Human Insecurities Unpacking AI Anxiety and Human Insecurities The rapid advancement of artificial intelligence has sparked excitement and trepidation in equal measure. As AI-powered systems permeate our daily lives—from recommendation algorithms to autonomous vehicles—many people grapple with a deep-seated unease. Far from being a purely technological debate, our fear of AI mirrors the most profound human insecurities: concerns about identity, control, and purpose in an ever-evolving world....

Humans instantly understand stories, emotions, and context. AI, on the other hand, learns through pixels, coordinates, labels, and patterns. The bridge between the two? High-quality annotation. Accurate data annotation helps AI interpret the world more intelligently powering smarter computer vision systems for autonomous vehicles, healthcare, retail, surveillance, and beyond. #ArtificialIntelligence #ComputerVision #DataAnnotation #MachineLearning #DeepLearning #AIInnovation #VisionAI #AITrainingData #TechInnovation #Automation #SmartAI #VaidikAI #FutureOfAI #vaidik

A fleet of robotaxis just exposed the real AI problem. It is not autonomy. It is judgment. A group of autonomous vehicles reportedly became stuck at a San Francisco intersection. No collision. No injuries. No system meltdown. Just a coordination failure. Each vehicle followed protocol. Each avoided risk. Each waited for the correct opening. Together, they created a standstill that required human help. A human driver likely solves this in seconds. Because cities do not run on logic alone. They run on negotiation. The small wave. The inch forward. The shared signal. The judgment call. The controlled ambiguity that keeps traffic moving. We have spent billions teaching machines to detect pedestrians, read signs, and stay in lanes. But the harder challenge is teaching systems contextual judgment. When to wait. When to move. When perfect compliance creates the wrong outcome. This is not just a robotaxi problem. It is an AI systems problem. Operators execute protocol. Orchestrators understand when protocol becomes the bottleneck. The future will not be run by systems that follow instructions perfectly. It will be run by systems that know when rigid compliance creates failure. If you are defining your AI strategy, let's talk. I help leaders design operating models that scale intelligent workflows across the enterprise. Save 💾 React 👍 Share ♻️ Follow 🔔

Well, this cleared eventually. But this is the best analogy for having a mechanism to communicate between agents. You have multiple agents all going to the same place, each one of them, trying to stay out of the way of the other agents, you need to have a protocol, which allows them to have that conversation about who's going to go where who's going to do what. This is true and any agentic situation, not just Robo taxis. Other than that, it's pretty funny.